/mlmodelscope

MLModelScope is an open source, extensible, and customizable platform to facilitate evaluation and measurement of ML models within AI pipelines.

Primary LanguageJavaScriptUniversity of Illinois/NCSA Open Source LicenseNCSA

MLModelScope Frontend

Getting Started

Prerequisites

  • Node.js (== 14.21.3 is the confirmed working version)
    • It is recommended to use nvm to manage your Node.js versions across multiple projects
  • Docker (latest version) and Docker Compose (latest version)
    • Docker is used to run the backend services (e.g. database, MLModelScope server) locally
    • Run the compose file found in the corresponding API repository in order to start the backend services for use in local development (https://github.com/c3sr/mlmodelscope-api)

First-time setup

  1. Install dependencies

    npm install

  2. Copy .env.example to .env. The default values in .env.example should be suitable for local development.

    cp .env.example .env

Running the app

In order to run the app completely, make sure the backend services are running (see Prerequisites above), then run the following command:

npm start

This will start the frontend app in development mode. Open http://localhost:3000 to view it in the browser.

Running storybook

Storybook is a tool for developing UI components in isolation. It is useful for developing and testing components. Additionally, there is a storybook deployment that is automatically updated when the develop branch is updated. The storybook deployment can be found here.

To run storybook locally, run the following command:

npm run storybook

This will start storybook in development mode. Open http://localhost:6006 to view it in the browser.

Running tests

To run the unit tests inside the terminal, run the following command:

npm test

In addition, most IDEs have a way to run tests from within the IDE. Refer to your IDE's documentation for more information on how to run Jest tests.

Project Structure

The project is structured as follows:

  • src/ - Contains the source code for the frontend app
    • components/ - Where the components for each of the pages reside. Subfolders within this folder are grouped by individual pages that are displayed on the website. Note that the React component (.js[x]), styles (.scss), and associated React hooks (use[*].js) will always be found within the same directory.
    • resources/ - Contains the resources (e.g. images, icons, etc.) used by the frontend app
    • /router/ - Contains page routes. Any new pages to the website are added through the routes array in router/config.js
    • /routes/ - Contains any wrapper components for specific page routes (for example, different wrappers for the model listing page for whether it is in "add mode" or "list mode"
    • /helpers/ - Contains helper functions and classes that are used throughout the frontend app
  • /public/ - Contains the public assets for the frontend app (e.g. favicon, index.html, etc.)
  • /.storybook/ - Contains the configuration for storybook
  • /scripts/ - Contains scripts that are used for development and deployment

Recommended Workflow for creating new Tasks (WIP)

  • Add the task name to TaskIDs.js. Use camelcase for the variable and snakecase for the string

  • Create a new Task in Task.js

    • Search svgrepo.com for a suitable icon and add it to the icons folder (see below for more instructions)
    • Add the new task to:
      • getStaticTasks
      • getDefaultModel
        • Open DefaultModels.js and copy/paste one of the existing models (editing where appropriate)
      • getSampleOutput
        • Return the Test[Task Name]Output that you created in testData (see next bullet point)

  • Make a new directory in Outputs and create:

    • testData subdirectory
      • testTaskNameOutput.js
    • TaskNameOutput.js
    • TaskName.stories.js
    • TaskName.scss

  • Create a new story for the task in QuickInput.stories.js

    • For the sample inputs, you'll create them below. Come back and import them here once you have.

  • If the task requires new input type(s), you will need to add those:

    • Add the new input type to TaskInputTypes and QuickInputType
    • Open useQuickInputControl.js
      • Update getInputTabType and getUploadTabType with the new QuickInputType
      • Create a new [InputType]InputTab file
        • You can look at the closed/merged Pull Requests for tasks like Table Editing or Audio to Text for examples of new Input Tabs.
      • You can temporarily switch the default state for selectedTab to the index of your new tab, so that you don't need to repeatedly switch tabs through page refreshes
    • Update SampleInputsTab:
      • Open sampleImages.js and add Sample[TaskName]Inputs. If your task is using useMultiInput then you will need to make a parent array, and then for each input type another array of sample input objects.
      • Go back to QuickInput.stories.js and update the sample inputs
      • Add the new input type to makeSampleInput
      • Create a makeSample[input type]Input function
      • Add the new input type with appropriate text to makeTaskTitle
      • Update the SampleInputsTab.scss file with styling for what the selected/unselected states of the new input type should look like
    • Update UploadInputsTab as necessary
      • If the user is able to upload files for this task, add the task ID to getAllowedFileTypes in UppyFileTypeCheckerPlugin so that Uppy will only allow the correct file types to be selected/sent to the server
    • Update URLInputsTab as necessary
      • Be sure to read the IMPORTANT comment before editing the inputs
        • There are multiple paths in the render function depending on if your input is a useMultipleInput, multiple, etc. - if you change one, you will likely need to update all of them.

  • Add the new task to ModelDetailPage in getSampleInputs and getInputType

  • If you haven't already done so, create the various Output files above, then:

    • Update QuickOutput with your new Output
      • In the makeOutput function, add a new case statement for your new Output
        • Generally you will only need to pass through the trialOutput and onBackClicked but if you need to be able to re-run the trial on the same page (for example with Conversations) then you would also add a prop here for runTrial.

  • Additional Notes:

    • To test the upload dashboard in storybook, open useUploadInputControl and:
      • Find the text COMMENT THIS OUT BEFORE COMMITTING and uncomment it.
      • Find the text UNCOMMENT THIS BEFORE COMMITTING and comment it out.
      • Be sure to reset these when you are finished. These changes are necessary to avoid making a call to S3 (which will not work) as well as to make a fake uploadURL so that our code will continue to execute as if the S3 call was successful.
    • To see what your currently-selected inputs, and the current state of the data that will be sent to the API (prior to clicking the "Run Model" button), go to useQuickInputControl.js and uncomment the useEffect with console.logs in it.
    • To check the array of inputs that you are submitting to the API, add an onRunModelClicked function to your task in QuickInput.stories.js (it takes the inputs as a param). Because of how Storybook works, the component you're building won't be passed the real method, but you can make a mock in the stories file to test behavior.
      • See TextConversationOutput.stories.js for an example of how to test api requests in this way

Adding new Task icons

  • Go to svgrepo.com, search for a suitable icon, download it, and drag the file into the src/resources/icons directory
  • Rename the file to follow the icon-taskNameInCamelCase.svg structure that all of the other Task icons use
  • Import the file to Task.js and finish Task setup as usual.
  • You can view the icon in the Home -> Splash Page Storybook.
  • After deploying to Staging (or locally if you have the React app running), the icon is also visible on the All Models page, in the "Filter Models by Tasks" menu on the left.
  • The icon colors should be white and blue. If you notice black lines, you'll need to use css to adjust those.
    • Open the svg file, and if the file isn't already formatted, from the main VS Code menu, select View and then Command Pallet, and then type Format document into the search bar (you may need to install XML Tools)
    • Now open the Header.scss and _FilterGroup.scss files and find the related styling (Search for this text: // Styling for .svg Task icons to make sure they are $white and $azul)
    • Compare the xml tags against the styling here, and update as necessary

Recommended Workflow for creating new components

Create a new folder in src/components/ with the name of the component, and create the following files as needed:

  • [ComponentName].jsx - The main React component
  • [ComponentName].scss - The styles for the component
    • Note that any new stylesheets should also be imported into src/Stopgap.scss
  • use[*].js - Any React hooks that are used by the component
  • [ComponentName].stories.jsx - The storybook stories for the component
  • [ComponentName].test.jsx - The unit tests for the component
  • use[*].test.js - The unit tests for the React hooks

Creating multi-image sample images:

  • Some tasks, such as ImageTo3D, use a collection of images rather than a single image on the sample tab
  • In order to easily display and select them as a group while preserving current functionality, we can use an image-editing tool to combine the images. You are welcome to use any tool you wish, but here are the steps to do so with Figma (free to sign up):
  1. Drag any images you want from your folder to the Figma canvas & line them up in a row.
  2. Select the "square" tool from the upper-left corner, and draw a square on the canvas, around one of the images. Then, on the right side menu, click the - button in the Fill row. This should make your square transparent except for its outer border.
  3. Copy and paste the square until you have one for each image, and move them to frame each image (to give the appearance that the images are separate, rather than one large image).
  4. Click on the canvas and drag to select all of the images+squares, then right-click on them and select Group selection to combine them.
  • On the right menu you'll also see the option to adjust the dimensions of your new Group. You can shrink them down to a smaller number, but be sure to click the Constrain Proportions button to link X and Y and the dimensions the same (it is the small button next to the W and H values).
  1. At the bottom of the right menu, you'll see the Export option. Use that to export your image as a .png, and add it to the project.
  2. Update the tasksWithLargeImages array in SampleInputsTab.js with the task that you are adding.

Writing tests for UI components

Note: Enzyme is no longer supported by React This project uses Enzyme for testing React components. Enzyme is a testing utility for React that makes it easier to test React components. It is recommended to use Enzyme's shallow rendering API to test components. Shallow rendering is useful for testing components in isolation, and it is recommended to use shallow rendering for unit tests. For more information on shallow rendering, refer to the Enzyme documentation. Note that due to some shortcomings in Enzyme's ability to test React hooks, it is recommended to use the React Testing Library for testing React hooks.

Project Wiki

https://wiki.mlmodelscope.org